Citrus black spot detection using hyperspectral image analysis

Abstract

A recently discovered fungal disease called citrus black spot, is threatening the Florida citrus industry.  The fungal disease, which causes cosmetic lesions on the rind of the fruit and can cause a tree to drop its fruit prematurely, could possibly lead to a ban on sales of fresh Florida citrus in other citrus-producing states.  The objective of this research is to develop a multispectral imaging algorithm to detect citrus black spots based on hyperspectral image data.  Hyperspectral images of citrus fruits (Valencias) were collected in the wavelength range of 480 nm to 950 nm.  Five surface conditions were examined, citrus black spot, greasy spot, melanose, wind scar, and normal one.  The first part of the image analysis determined the optimal wavelengths using correlation analysis based on the wavelength ratio (l₁/l₂) and wavelength difference (l₁ - l₂).  Four wavelengths were identified, 493 nm, 629 nm, 713 nm, and 781 nm.  In the second part, pattern recognition approaches namely linear discriminant classifier and artificial neural networks were developed using the four selected wavelengths as the input.  Both pattern recognition approaches had an overall accuracy of 92%.  The detection accuracy was improved to 96% by using the NDVI band ratio method of 713 nm and 781 nm.  The multispectral image algorithm developed in this study haspotential to be adopted by a real-time multispectral imaging system for citrus black spot detection.

Keywords: activation energy, effective diffusivity, foam-mat drying, foam characteristics, modeling, Shrimp